American English AI Voices

Accent range inside American English

General American is a broadcast standard rather than a regional accent, and it is what most of these voices speak. That is the right default for narration and for characters whose region is not part of the story. It is the wrong default when place matters — a story set somewhere specific with an unplaceable narrator reads as generic.

There are no regional American voices, and no accent control either: you choose a library character and the voice follows from it. Where region is load-bearing, the honest options are to signal it through vocabulary and rhythm in the writing, or to record those characters yourself.

Numbers, dates and abbreviations

A performance directed as American reads dates and abbreviations the way an American would — month-day-year, and the American expansion of an abbreviation. If your text mixes conventions — a British-set story with American narration, say — check the numbers, because that is where mismatches show up first and are least forgivable.

There is no pronunciation dictionary, so a word that comes out wrong has to be handled in the script itself. Respelling it the way it sounds is crude and it works; audition the line, not the word on its own, because the surrounding rhythm often fixes what looks broken in isolation.

A scene, cast line by line

Six lines from a kitchen, three speakers, all of them characters from this page. The names in the script are the library characters themselves, which is the simplest way to work: whatever name sits before the colon becomes a character in the project, and you point it at a library character from there. Cass is seventeen and already packed; Hazel is her mother; Orrin is Hazel's father. After each line is what we would cast and how we would direct it.

CASS: “I already filled the tank. I'm leaving at five, with or without a blessing.” — Cass is the teenager in the library and the obvious casting. The line wants excited rather than angry, because she has already won and knows it. Anger here makes her sound like she expects a fight, which flattens the five lines that follow.

HAZEL: “You are leaving at five with a blessing. I just haven't decided whose.” — Hazel is written as patient and then suddenly not, so her default emotion is the thing to get right: set it to neutral and let the writing carry the edge. Every character has a default emotion on every plan, and this is a part where the default does almost all the work.

ORRIN: “Mine's free. Always has been.” — Orrin is the old man on the porch, and the two short sentences are the whole performance. Keep him neutral and, on Plus or Pro, take the line to 0.9×. The slowest thing in a scene tends to read as the most certain thing in it.

HAZEL: “Dad, don't. She's seventeen.” — this is the line to mark. Angry if your Hazel is the kind who raises her voice, sad if she is not; both are among the seven presets and both are defensible readings of the same words. Line-by-line emotion is a Pro control, so on other plans you would decide which of the two the whole part is and let every line of hers follow it.

ORRIN: “So were you, the year you took my truck to Reno.” — happy, and smaller than you think. The joke only lands if he is not enjoying it too loudly. Audition this one twice: a second take of the same line is never identical, and on a dry line the difference between two takes is wider than anywhere else.

CASS: “…You took a truck to Reno?” — back to Cass, and the only genuinely hard line here, because the performance has to turn in the middle of it. Try excited. If the turn does not arrive, the fix is the text rather than the direction: split it into two lines for Cass, and the short gap between two lines from the same character does the work.

Three speakers, three of the ten base voices, which is the shape most scenes have. Worth noticing what never appeared: nobody wrote an accent, because inside American English there is nothing to write, and nobody wrote a stage direction into the middle of a line. A parenthetical at the start of a line, or on a line of its own, becomes direction for the performance; a parenthetical inside a line is read aloud. The rest of the direction lives in the emotion you set and in the words you keep.

What to listen for before you export

Exporting spends download credits and re-exporting the same project spends them again, so the listening pass is worth doing once, properly, on the assembled scene rather than line by line. Six things go wrong in American English more often than anything else.

Dates, times and money. Listen for the order and for the words the reader supplies: 9/11/2026 should come out September eleventh, and 5:00 should not come out five colon zero zero. Anything that has to be heard exactly is safer written as words — five o'clock, nine dollars fifty — because the text is the only lever there is.

Acronyms and initialisms. NASA is a word, FBI is three letters, and nothing in the spelling says which. Listen for the one your audience says. Where it comes out wrong, space the letters (F B I) or spell the word (Nasa) and audition again.

Homographs. Read, lead, live, bow, tear, wound. A wrong pick is a wrong word rather than an odd inflection, and short lines are where you hear it, because there is no rhythm for it to hide behind. Rewriting the sentence so the tense is unambiguous fixes more of them than respelling does.

Proper nouns. Invented names, surnames and small American towns are where it fails. Listen to the protagonist's name three times in a row; if it wobbles, respell it phonetically in the script. The respelling also appears in the SRT subtitles on Plus and Pro, where you can put it back in your editor.

Long lines. Anything past about thirty words flattens somewhere in the middle. Listen for the point where the intonation stops moving, then break the line into two lines for the same character — it costs nothing and puts a short pause where the thought turns.

Two takes of one line. Regenerate a line you were happy with and play both. They will not match, because every take is a fresh performance, and hearing the range once tells you whether the take you kept is the good one or just the first one.